Skip to main content
Digital ArchaeologyMethod BreakthroughAug 26, 2026, 4:55 PM· 5 min read· in science

AI Solves 'Impossible' Archaeological Challenge by Unwrapping First Complete Herculaneum Scroll

Using high-resolution X-rays and machine learning, researchers have read an entire carbonized scroll from the ancient Herculaneum library from end to end without ever physically opening it.

By Viktoria Sokolova

Papyrologists and Classicists 40%Computer Scientists 35%Heritage Preservationists 25%
Papyrologists and Classicists
Scholars focused on the translation and historical context of the newly recovered philosophical texts.
Computer Scientists
Researchers focused on the machine learning architecture and the physical detection of ink without generative hallucination.
Heritage Preservationists
Conservators focused on the non-destructive nature of the technology and the preservation of the physical artifacts.

What we don’t know

  • The exact author of the Stoic treatise in PHerc. 1667 remains unconfirmed, though it names Aristocreon.
  • It is unclear how long it will take to fully automate the 3D segmentation process to scale this to all 600 remaining scrolls.
  • Archaeologists still debate whether the excavated Villa of the Papyri represents the main library or if a larger archive remains buried.

For the first time in nearly two millennia, an entire carbonized scroll from the ancient Roman town of Herculaneum has been read from beginning to end without ever being physically opened. Researchers combining high-energy X-ray microtomography with machine learning have successfully extracted roughly 1.4 meters of continuous ancient Greek text from a sealed, charred lump of papyrus known as PHerc. 1667. The recovered document, a 2nd-century BC Stoic philosophical treatise, marks the most significant milestone in digital archaeology to date, proving that the hundreds of remaining unread scrolls can be deciphered without destroying them.

The breakthrough, announced by the Vesuvius Challenge and researchers at the University of Kentucky, solves a fundamental physical paradox that has plagued classicists for centuries. When Mount Vesuvius erupted in 79 AD, the intense heat carbonized the library of the Villa of the Papyri. The scrolls survived only because they were baked into brittle charcoal. For over 200 years, attempting to unroll them physically meant watching them crumble into dust. The new computational pipeline separates physical access from textual access: the artifact remains safely sealed in a vault, while its contents are extracted mathematically.

The subject of this milestone, PHerc. 1667, is a battered survivor. Historical records from 1782 describe it as a compressed but largely intact object. However, subsequent physical attempts to pry it open—most recently in the 1980s—caused the outer layers to flake off and disintegrate. What remains today is only the compact inner core, measuring just 8 centimeters tall and 2 centimeters wide. Yet, from this heavily damaged fragment, the digital pipeline recovered 22 continuous columns of text.

Despite severe damage from past physical unwrapping attempts, the surviving core of PHerc. 1667 yielded extensive continuous text.

Extracting text from a fused lump of carbon requires a three-step mechanism, beginning with particle accelerators. Because the scrolls are dense and tightly wound, standard medical X-rays cannot penetrate them with enough resolution. Instead, researchers used phase-contrast X-ray microtomography at the European Synchrotron Radiation Facility (ESRF) in Grenoble, France, and the Diamond Light Source in the UK. These high-energy beams map the internal 3D geometry of the scroll at a microscopic scale, capturing the exact position of every warped, compressed layer of papyrus.[1][2]

The second step is computational geometry. The X-ray data produces a massive 3D volume of voxels (3D pixels). Software algorithms must trace the spiraled layers of papyrus through this volume, reconstructing them as explicit three-dimensional surfaces. Once a continuous sheet is mapped in 3D space, the software mathematically flattens it into a 2D plane, effectively "unrolling" the scroll on a screen while the physical object remains untouched.

The X-ray data produces a massive 3D volume of voxels (3D pixels).

The final and most difficult step is finding the ink. The Herculaneum writers used a carbon-based ink made of soot and water. Because the ink is made of the exact same material as the carbonized papyrus it sits on, it does not show up as a bright contrast in X-ray scans. To solve this, researchers trained machine-learning models to detect microscopic structural changes in the papyrus fibers—subtle variations in thickness or texture—caused by the application of the ink before the scroll was burned.[2]

The computational pipeline separates physical access from textual access, mathematically unrolling the scroll's geometry before detecting the ink.

The evidence strength of this method is rigorously constrained to prevent hallucinations. The machine-learning components do not know ancient Greek, nor do they use large language models to guess missing words based on context. Their sole function is to amplify weak physical evidence of ink in the flattened 2D images. The output is a visual map of ink traces, which human papyrologists then interpret and translate. This ensures that the recovered text is a direct physical measurement of the artifact, not a generative AI prediction.[2]

The resulting text from PHerc. 1667 is a previously unknown philosophical treatise on ethics. Papyrologists have identified it as a Stoic work discussing human nature, impulse, and moral progress. The final preserved column explicitly names Aristocreon, a nephew and disciple of the prominent Stoic philosopher Chrysippus. This internal evidence, combined with the language and themes, securely dates the composition to the 2nd century BC.

The ability to read 22 continuous columns represents a transformational shift in how scholars interact with the Herculaneum library. Previously, virtual unwrapping efforts had only recovered isolated words or small patches of text, as seen in the Vesuvius Challenge's 2024 Grand Prize, which extracted roughly 2,000 characters. Now, scholars can follow sustained philosophical arguments across multiple pages, treating the scrolls as complete literary works rather than fragmented archaeological puzzles.

The machine learning models do not guess words; they strictly amplify the physical evidence of carbon ink on the papyrus.

The pipeline is already scaling to other texts. Alongside the complete reading of PHerc. 1667, the team announced the recovery of more than 70 columns of text from PHerc. 172, housed at Oxford's Bodleian Library, identifying it as the Epicurean philosopher Philodemus's "On Vices, Book 1." They also recovered the title of PHerc. 139, identifying it as Philodemus's "On Gods, Book 8," proving that the work extended to at least eight volumes.[1]

Despite the success, significant limitations remain. The virtual unwrapping process is still highly computationally expensive and requires intense manual intervention, particularly during the "segmentation" phase where the 3D layers of papyrus are traced. The Vesuvius Challenge has published a master plan aimed at automating this surface extraction to reduce the manual labor required for each scroll. At current processing rates, completing the full library would take several years.

The stakes for automating this process are immense. More than 600 unopened scrolls remain in the Naples collection, representing the only intact library to survive from the classical world. With a proven, non-destructive method now established, researchers are poised to unlock an unprecedented volume of ancient thought, potentially recovering lost works of poetry, history, and philosophy that have been silent since the ash fell in 79 AD.

Key points

  1. Researchers used AI and X-ray microtomography to read an entire Herculaneum scroll without physically opening it.
  2. The scroll, PHerc. 1667, yielded roughly 1.4 meters of continuous ancient Greek text across 22 columns.
  3. The text is a previously unknown 2nd-century BC Stoic philosophical treatise on ethics and human nature.
  4. The machine learning models strictly detect physical ink traces and do not use language models to guess words.
  5. The breakthrough proves that the remaining ~600 fragile, carbonized scrolls can be read non-destructively.
1.4 meters
Continuous text recovered
22
Columns of ancient Greek read
8 cm
Surviving height of the scroll core
~600
Unopened scrolls remaining

Sources

Source coverage

2 outlets

3 viewpoints surfaced

Papyrologists and Classicists 40%Computer Scientists 35%Heritage Preservationists 25%
  1. [1]Diamond Light SourceHeritage Preservationists

    Vesuvius Challenge breakthrough reveals new texts from ancient world

    Read on Diamond Light Source
  2. [2]ExplainXComputer Scientists

    Vesuvius Challenge Reads an Entire Herculaneum Scroll — First Time in 2,000 Years

    Read on ExplainX

Comments

Stay informed

Every angle. Every day.

Get science stories with full source coverage and perspective breakdowns delivered to your inbox.